Toro
Diamond Member
First, I'm not making any predictions about the election whatsoever. I'm not saying Biden or Trump is going to win. Nor am I saying that that this theory isn't true. I just want to see if there is any empirical data to suggest that this "Shy Trump Voter" theory is statistically true.
The theory is that many Trump voters don't want to tell pollsters that they are really voting for Trump. The theory psotulates that social disapproval leads people not to be truthful that they are supporting the President.
If this were the case - that voters were telling pollsters one thing but doing the other - it should be pretty simple to see in the data. On a net basis, what that means is that Trump's actual number of votes should materially differ from what pollsters were estimating Trump's vote to be.
I looked at 13 states from the 2016 election - six that most pundits consider swing states, and seven that pundits think will not flip but could be in play. The data is from RCP.
The first column is the estimate of Trump's vote by the polling firms. The second column is Trump's actual votes. As you can see, in all but one case, Trump's vote was actually higher than what pollsters were estimating.
But that was true for the most part for Clinton too.
Why would that happen? Why would both candidate's actual vote tallies be higher than what was being shown in the polls? Because people who were undecided made their decision on election day. And they voted mostly for Trump.
But assuming that the differential between Trump's actual votes and polling numbers were all because of shy voters (they weren't), and not because people were genuinely undecided until they stepped into the polling booth, it appears that there isn't much statistical significant evidence of this theory. Perhaps except for one very notable, and important, exception.
The average between Trump's votes and polling numbers of the 13 states is 3.2%. Most state polls have a margin of error of 4% or greater, so this average is likely not statistically significant. In fact, the differences are below 4% in 10 of the 13 states.
However, it is above 4% in three critical states - IA OH and WI - all states Trump flipped, and were critical in him winning. So perhaps the Shy Voter Theory only applies to the (critically important) Midwest.
There is no evidence that people said they were voting for Clinton then voting for Trump. Or if there was, it wouldn't matter because what matters is the net differential between what the polls were telling us and what actually happened. IOW, there may have been pollsters who were "Shy Hillary Voters," and saying they were voting for Trump and actually voting for Clinton. But that's unlikely. Thus, most "Shy Trump Voters" were probably saying they were undecided or not saying anything.
The problem this time around for Trump is that there are fewer people who are undecided. These numbers come from both RCP and 538. Those who are saying they are undecided is down by about a third from last election.
This is especially a problem for Trump because he is further behind Biden than he was behind Clinton. In all 13 states, Biden is out-polling Clinton.
And if you look at the gap between the lead Biden has and the number of undecideds/other, i.e. the ground Trump has to make up on Biden, it looks pretty daunting.
The higher the number, the better it is for Trump. Anything with a zero or a negative number means there is no runway for Trump to catch up based on current polling. In PA, it's virtually zero. If this is true, then Biden would flip MI PA and WI and win the election.
Now does this mean Biden is certainly going to win? No, of course not. Of course Trump can win. Something may happen between now and the election. Polls might be off. The pandemic may affect voting. Maybe Biden really does have Alzheimer's and he shows it. Who knows?
But this is a different election than 2016, and it appears Trump has a higher hill to climb.
The theory is that many Trump voters don't want to tell pollsters that they are really voting for Trump. The theory psotulates that social disapproval leads people not to be truthful that they are supporting the President.
If this were the case - that voters were telling pollsters one thing but doing the other - it should be pretty simple to see in the data. On a net basis, what that means is that Trump's actual number of votes should materially differ from what pollsters were estimating Trump's vote to be.
I looked at 13 states from the 2016 election - six that most pundits consider swing states, and seven that pundits think will not flip but could be in play. The data is from RCP.
The first column is the estimate of Trump's vote by the polling firms. The second column is Trump's actual votes. As you can see, in all but one case, Trump's vote was actually higher than what pollsters were estimating.
But that was true for the most part for Clinton too.
Why would that happen? Why would both candidate's actual vote tallies be higher than what was being shown in the polls? Because people who were undecided made their decision on election day. And they voted mostly for Trump.
But assuming that the differential between Trump's actual votes and polling numbers were all because of shy voters (they weren't), and not because people were genuinely undecided until they stepped into the polling booth, it appears that there isn't much statistical significant evidence of this theory. Perhaps except for one very notable, and important, exception.
The average between Trump's votes and polling numbers of the 13 states is 3.2%. Most state polls have a margin of error of 4% or greater, so this average is likely not statistically significant. In fact, the differences are below 4% in 10 of the 13 states.
However, it is above 4% in three critical states - IA OH and WI - all states Trump flipped, and were critical in him winning. So perhaps the Shy Voter Theory only applies to the (critically important) Midwest.
There is no evidence that people said they were voting for Clinton then voting for Trump. Or if there was, it wouldn't matter because what matters is the net differential between what the polls were telling us and what actually happened. IOW, there may have been pollsters who were "Shy Hillary Voters," and saying they were voting for Trump and actually voting for Clinton. But that's unlikely. Thus, most "Shy Trump Voters" were probably saying they were undecided or not saying anything.
The problem this time around for Trump is that there are fewer people who are undecided. These numbers come from both RCP and 538. Those who are saying they are undecided is down by about a third from last election.
This is especially a problem for Trump because he is further behind Biden than he was behind Clinton. In all 13 states, Biden is out-polling Clinton.
And if you look at the gap between the lead Biden has and the number of undecideds/other, i.e. the ground Trump has to make up on Biden, it looks pretty daunting.
The higher the number, the better it is for Trump. Anything with a zero or a negative number means there is no runway for Trump to catch up based on current polling. In PA, it's virtually zero. If this is true, then Biden would flip MI PA and WI and win the election.
Now does this mean Biden is certainly going to win? No, of course not. Of course Trump can win. Something may happen between now and the election. Polls might be off. The pandemic may affect voting. Maybe Biden really does have Alzheimer's and he shows it. Who knows?
But this is a different election than 2016, and it appears Trump has a higher hill to climb.